Motion planning plays a significant role in enabling advances of autonomous vehicles in saving lives and improving traffic efficiency. In a predictive motion-planning strategy, the ego vehicle predicts the motion of surrounding vehicles and uses these predictions to plan collision-free reference trajectories. In dynamic multi-vehicle traffic environments, a key research question is how to consider vehicle-to-vehicle interactions and motion uncertainties of the surrounding vehicles in the motion planner to achieve resilient motion planning of the autonomous ego vehicle.  This Licentiate Thesis proposes a model predictive control (MPC)-based approach to achieve safe motion planning in uncertain and dynamic multi-vehicle driving environments. MPC has been widely applied for the motion planning of autonomous vehicles. However, designing resilient MPC-based motion planners that consider interactions and uncertainties of surrounding vehicles remains an open and challenging problem, which is the primary motivation for the research presented in this thesis.  This thesis makes several contributions toward solving the interaction and uncertainty-aware motion-planning problems. The first contribution is an MPC, which is called interaction-aware moving target MPC. It is designed based on the combination of an interaction-aware motion-prediction model and time-varying reference targets of the optimal control problem for proactive and non-local trajectory planning in multi-vehicle dynamic scenarios.  In the second contribution, the proposed MPC is extended to account for the multi-modal motion uncertainties of surrounding vehicles, including the maneuver and trajectory uncertainties, which are predicted by combining an interaction-aware motion-prediction model and a data-driven approach. Based on the modeling of uncertainties, a safety-awareness parameter is included in the design to compute the obstacle occupancy for achieving a trade-off between the performance and robustness of the MPC planner. The efficiency of the method is illustrated in challenging highway-driving simulation scenarios and a driving scenario from a recorded traffic dataset.  The third contribution of this thesis is quantifying the motion uncertainty of surrounding obstacles to reduce the conservativeness of the motion planner while pursuing robustness. To this end, a robust motion-planning method is designed for robotic systems based on uncertainty quantification of surrounding obstacles. The proposed MPC is called risk-aware robust MPC, as the risk of robustness reduction through uncertainty quantification is analyzed. Simulations in highway merging scenarios of an autonomous vehicle with uncertain surrounding vehicles show that the approach is less conservative than a conventional robust MPC and more robust than a deterministic MPC. / Rörelseplanering spelar en betydande roll för att möjliggöra framsteg inom autonoma fordon med potential att rädda liv genom att undvika olyckor och förbättra trafikeffektiviteten. I en prediktiv rörelseplaneringsstrategi predikterar det egna fordonet rörelsen hos omgivande fordon och använder dessa prediktioner för att planera en säker trajektoria. I dynamiska trafiksituationer med multipla omgivande fordon är en central forsknings-fråga hur man ska ta hänsyn till de omgivande fordonens interaktioner och rörelseosäkerheter för att åstadkomma en robust rörelseplanering. Den här licentiatavhandlingen föreslår en modellprediktiv reglerings-ansats (MPC) för rörelseplanering i osäkra och dynamiska flerfordonsmiljöer. Robust och säker modellprediktiv regleringsbaserad rörelseplanering som tar hänsyn till interaktioner och osäkerheter hos rörelsen för omgivande fordon är ett öppet och utmanande problem, vilket är den primära motiveringen för den forskning som presenteras i denna avhandling. Modellprediktiv reglering (MPC) är en vanlig ansats för rörelseplanering för autonoma fordon. Denna avhandling presenterar metoder som är steg mot att lösa rörelseplaneringsproblemet där interaktion mellan fordon och osäkerhet i rörelser för omgivande fordon beaktas. Det första bidraget fokuserar på interaktionen mellan omgivande fordon. En modellprediktiv regulator har utvecklats baserat på en modell för hur omgivande fordon interagerar och påverkar varandras beteende. Denna modell integreras sedan som tidsvarierande referensmål för det optimala styrningsproblemet vilket ger en förutseende och robust planering för det egna fordonet. I det andra bidraget utökas den föreslagna MPC-metoden för att ta hänsyn till de multimodala rörelseosäkerheterna hos omgivande fordon; det finns en osäkerhet i vad de omgivande fordonen kommer göra härnäst och det är osäkert hur de kommer genomföra det. Baserat på en modellering av osäkerheterna, en delvis datadriven ansats, inkluderas en säkerhetsparameter i regulatorn som möjliggör en avvägning mellan prestanda och robusthet hos MPC-planeraren. Den tredje bidraget i avhandlingen är nya metoder för att kvantifiera rörelseosäkerheten hos omgivande fordon och att använda denna för robust planering, utan att det egna fordonet blir för konservativ i sitt agerande. Den föreslagna ansatsen bygger på robust MPC där en riskmedvetenhet introduceras. Simuleringar av motorvägskörning med omgivande fordon med rörelseosäkerhet visar att metoden är mindre konservativ än en konventionell robust MPC och mer robust än en deterministisk MPC. / <p><strong>Funding:</strong> This research was supported by the Strategic Research Area at Linköping-Lund in Information Technology (ELLIIT).</p>

Physics-based models of dynamical systems can take a lot of time and be hard to derive, and there will always be some effect that was not added to the calculations, like aerodynamic-, gyroscopic- or frictional- effects. Calculating all these effects takes time and a lot of knowledge of the system dynamics. There are many different ways to implement different methods to speed up the process of creating the control policy. What is the simplest way to create a control policy and how does it control the system? Neural networks is a promising approach, where there are two different methods. First, by using the mathematical structure of a neural network a model of the system can be derived, and then a simple control policy is used. Second, Reinforcement learning is where the control policy is learned. These two are compared to a baseline model where the model of the system is derived from the physical description of the system. First, the model is calculated by the system dynamics with classical mechanics that describes the mathematical description of a physics-based system. Then the machine-learning approach of using a neural network to learn and describe the system is implemented. Lastly, the Reinforcement learning method is made and compared to the other models. The models had all their own differences in performance. The controllers based on the physics-based model were good in a small region around the equilibrium and it took a long time to derive. The neural network models were more general and easier to implement but were more unstable, they showed the problems with data collection for training the model, here several approaches could be used to improve the model and patch the problems seen. Lastly, the reinforcement controller worked well but from a control theory perspective, it is very hard to prove the stability of the controller.

In this thesis, we study trajectory tracking and prediction-based control of underactuated unmanned aerial and surface vehicles.  In the first part of the thesis, we examine the trajectory tracking using prescribed performance control (PPC) assuming that the model parameters are unknown. Moreover, due to the underactuation the original PPC is redesigned to accommodate for the specifics of the considered underactuated systems. We prove the stability of the proposed control schemes and support it with numerical simulations on the quadrotor and boat models. Furthermore, we propose enhancements to kinodynamic motion-planning via funnel control (KDF) framework that are based on rapidly-exploring random tree (RRT) algorithm and B-splines to generate the smooth trajectories and track them with PPC. We conducted real-world experiments and tested the advantages of the proposed enhancements to KDF. The second part of the thesis is devoted to the rendezvous problem of autonomous landing of a quadrotor on a boat based on distributed model predictive control (MPC) algorithms. We propose an algorithm that assumes minimal exchange of information between the agents, which is the rendezvous location, and an update rule to maintain the recursive feasibility of the landing. Moreover, we present a convergence proof without enforcing the terminal set constraints.  Finally, we investigated a leader-follower framework and presented an algorithm for multiple follower agents to land autonomously on the landing platform attached to the leader. An agent is equipped with a trajectory predictor to handle the cases of communication loss and avoid the inter-agent collisions. The algorithm is tested in a simulation scenario with the described challenges and the numerical results support the theoretical findings. / I denna avhandling studerar vi banspårning och prediktionsbaserad styrning av underaktuerade obemannade luft- och ytfarkoster. I den första delen av avhandlingen undersöker vi banspårningen med hjälp av föreskriven prestationskontroll (PPC) förutsatt att modellparametrarna är okända. På grund av underaktueringen i systemen vi betraktar har den ursprungliga PPC:n dessutom designats om för att specifikationerna för dessa system. Vi bevisar att de föreslagna regulatorerna stabiliserar systemet och validerar dem med numeriska simuleringar på både quadrotor- och båtmodellen. Dessutom föreslår vi förbättringar av kinodynamisk rörelseplanering via ramverk för trattkontroll (KDF) som är baserade på algoritmen för snabbutforskande slumpmässiga träd (RRT) och B-splines för att generera släta banor och spåra dem med PPC. Vi genomförde fysikaliska experiment och validerade fördelarna med de föreslagna förbättringarna av KDF. Den andra delen av avhandlingen ägnas åt mötesproblemet med autonom landning av en quadrotor på en båt baserat på algoritmer för distribuerad modell-prediktiv styrning (MPC). Vi föreslår en algoritm som förutsätter ett minimalt utbyte av information mellan agenterna, nämligen mötesplatsen, och en uppdateringsregel för att upprätthålla den rekursiva genomförbarheten av landningen. Dessutom presenterar vi ett konvergensbevis utan att upprätthålla begränsningar i slutuppsättningen. Slutligen undersökte vi ett ledare-följare ramverk och presenterade en algoritm där flera följaragenter kan autonomt landa på en plattform som sitter fast i ledaren. En agent är utrustad med en banprediktor för att hantera fall av kommunikationsbortfall samt för att undvika kollision med andra agenter. Algoritmen testas i ett scenario med de beskrivna utmaningarna och de numeriska resultaten överensstämmer med de teoretiska resultaten. / <p>QC 20230412</p>

Automated driving systems are becoming more prevalent in the heavy vehicle transportation industry. Increased automation is presumed to have positive effects on the industry's safety, productivity, and environmental impact. One essential part of increased automation is the development of high-level functionalities such as motion planning and control algorithms.  This thesis proposes a new system structure for automated driving system implementations, using a simplified vehicle model with physically motivated constraints, referred to as a reference model, to give a promised behavior on future states of the vehicle. In the proposed system, there exists a high-level controller which utilizes the reference model to calculate a reference signal which corresponds to a realizable behavior. A low-level controller with better knowledge of the plant dynamics and direct access to the plant actuators is then used to realize the reference signal for the specific vehicle it is implemented in. In this thesis, the reference signal is chosen as the longitudinal acceleration of the vehicle. There are several advantages to such a system, the main one being the ability to re-use high-level functionalities between different vehicle types due to the usage of a standardized interface. This could decrease development time, as it would be possible to develop the high and low-level functionalities separately, and therefore work on both can be conducted simultaneously. These can later be merged in the implemented system. The work concerns the modeling of the longitudinal dynamics of vehicles. Models of varying complexities, in terms of available actuators are presented and evaluated to show the system's usability in the presence of parametric uncertainties as well as modeling errors. Model predictive control allocation is used and motivated for choosing which actuators to use for realizing a given reference signal. Model predictive control is also utilized as an application of the reference model where a velocity-following case is investigated. Static and dynamic constraints on longitudinal acceleration are derived for a general vehicle, and system identification of vehicle actuators' aggregate dynamics is discussed and implemented.

Decision-making is the mechanism of using available information to develop solutions to given problems by forming preferences, beliefs, or selecting courses of action amongst several alternatives. It is the main focus of a variety of scientific fields such as robotics, finances, and neuroscience. In this thesis, we study the mechanisms that generate behavior in diverse decision-making settings (the forward problem) and how their characteristics can explain observed behavior (the inverse problem). Both problems take a central role in current research due to the desire to understand the features of system behavior, many times under situations of risk and uncertainty. We study decision-making problems in the three following settings. In the first setting, we consider a decision-maker who forms a private belief (posterior distribution) on the state of the world by filtering private information. Estimating private beliefs is a way to understand what drives decisions. This forms a foundation for predicting, and counteracting against, future actions. In the setting of adversarial systems, we answer the problems of i) how can an adversary estimate the private belief of the decision-maker by observing its decisions (under two different scenarios), and ii) how can the decision-maker protect its private belief by confusing the adversary. We exemplify the applicability of our frameworks in regime-switching Markovian portfolio allocation. In the second setting we shift from an adversarial to a cooperative scenario. We consider a teacher-student framework similar to that used in learning from demonstration and transfer learning setups. An expert agent (teacher) knows the model of a system and wants to assist a learner agent (student) in performing identification for that system but cannot directly transfer its knowledge to the student. For example, the teacher's knowledge of the system might be abstract or the teacher and student might be employing different model classes, which renders the teacher's parameters uninformative to the student. We propose correctional learning as an approach where, in order to assist the student, the teacher can intercept the observations collected from the system and modify them to maximize the amount of information the student receives about the system. We obtain finite-sample results for correctional learning of binomial systems. In the third and final setting we shift our attention to cognitive science and decision-making of biological systems, to obtain insight about the intrinsic characteristics of these systems. We focus on time perception - how humans and animals perceive the passage of time, and solve the forward problem by designing a biologically-inspired decision-making framework that replicates the mechanisms responsible for time perception. We conclude that a simulated robot equipped with our framework is able to perceive time similarly to animals - when it comes to their intrinsic mechanisms of interpreting time and performing time-aware actions. We then focus on the inverse problem. Based on the empirical action probability distribution of the agent, we are able to estimate the parameters it uses for perceiving time. Our work shows promising results when it comes to drawing conclusions regarding some of the characteristics present in biological timing mechanisms. / <p>QC 20210521</p>

The goal of this project was to calculate how much wind power could be balanced with hydro power in our designated area consisting of five hydro power stations, four villages which consume power, possible locations for wind power and one connection to the national grid. To achieve this a simulation model was constructed in the GAMS software with the goal of achieving the maximum profit from the hydro power plants by considering electricity prices, inflow of water, the physical construction of the power plants and the time of year. When this was achieved, restriction for the maximum transmission load on the power grid was added as well as local wind power production as to simulate the implementation of new power sources on an old system and power grid. This would result in a maximum income in SEK as well as the most wind power which could be maintained and balanced by the designated system. This project shows how to find the optimal way to use hydro power and wind power as well as how the integration between different sources of electricity production could work, which is vital for a future powered by renewable energy and will help towards lowering emissions.

The Reliability-Centred Asset Maintenance (RCAM) is a structured approach to determine maintenance strategies for electric power system which is developed at KTH School of Electrical Engineering. RCAM focus on reliability aspects of the system and one of the main steps in RCAM in to modeling the relationship between reliability and the elect of maintenance for critical components within the system. The transformer has been identified as a critical component within a power system and in spring 2006 the Ph.D project “Life time modeling and management of transformers” was started within the RCAM group. The overall goal for the project is to develop a quantitative model for the lifetime distribution of a transformer with the final goal to implement the developed model into a maintenance planning. Dissolved Gas Analysis (DGA) is a widely used technique to estimate the condition of oil-immersed transformers. Incipient faults within the transformer may be detected by analyzing the gases which are dissolved in the transformer-oil. The objective of this thesis is mainly to analyze available data from DGA, and investigate if this kind of data may be useful in quantitative modeling of the transformers reliability. One conclusion from this work is that the difficulty in modeling the transformers reliability lies mainly in the limited availability of adequate data. Transformer is a reliable device and since the number of failures in critical in mathematical modeling of a component´s reliability it becomes very difficult to determine such models. Another aspect is the difficulty to draw conclusions about a transformer´s condition only from the DGA results. Although there are standards available for this purpose the DGA interpretation should also be based on other information about the particular transformed such as size, construction and operation circumstance. During this work no sources have been found from which the correlation between DGA data and probability for transformer failures could be estimated. For this reasons the proposed failure rate function in this work is based on several subjective assumptions and has not been possible to verify.

Stochastic programming is a subfield of mathematical programming concerned with optimization problems subjected to uncertainty. Many engineering problems with random elements can be accurately modeled as a stochastic program. In particular, decision problems associated with hydropower operations motivate the application of stochastic programming. When complex decision-support problems are considered, the corresponding stochastic programming models often grow too large to store and solve on a single computer. This clarifies the need for parallel approaches that could enable efficient treatment of large-scale stochastic programs in a distributed environment. In this thesis, we develop mathematical and computational tools in order to facilitate distributed stochastic programs that can be efficiently stored and solved. First, we present a software framework for stochastic programming implemented in the Julia language. A key feature of the framework is the support for distributing stochastic programs in memory. Moreover, the framework includes a large set of structure-exploiting algorithms for solving stochastic programming problems. These algorithms are based on the classical L-shaped and progressive-hedging algorithms and can run in parallel on distributed stochastic programs. The distributed performance of our software tools is improved by exploring algorithmic innovations and software patterns. We present the architecture of the framework and highlight key implementation details. Finally, we provide illustrative examples of stochastic programming functionality and benchmarks on large-scale problems. Then, we pursue further algorithmic improvements to the distributed L-shaped algorithm. Specifically, we consider the use of dynamic cut aggregation. We develop theoretical results on convergence and complexity and then showcase performance improvements in numerical experiments. We suggest several aggregation schemes that are based on parameterized selection rules. Before we perform large-scale experiments, the aggregation parameters are determined by a tuning procedure. In brief, cut aggregation can yield major performance improvements to L-shaped algorithms in distributed settings. Finally, we consider an application to hydropower operations. The day-ahead planning problem involves specifying optimal order volumes in a deregulated electricity market, without knowledge of the next-day market price, and then optimizing the hydropower production. We provide a detailed introduction to the day-ahead model and explain how we can implement it with our computational tools. This covers a complete procedure of gathering data, generating forecasts from the data, and finally formulating and solving a stochastic programming model of the day-ahead problem. Using a sample-based algorithm that internally relies on our structure-exploiting solvers, we obtain tight confidence intervals around the optimal solution of the day-ahead problem.

Critical infrastructures, such as the power grid and water distribution networks, are the backbone of our modern society. With the integration of computational devices and communication networks in critical infrastructures, they have become more efficient, but also more vulnerable to cyberattacks. Due to the underlying physical process, these cyberattacks can not only have a financial and ecological impact, but also cost human lives. Several reported cyberattacks on critical infrastructures show that it is vital to protect them from these attacks. Critical infrastructures typically rely on accurate sensor measurements for optimal performance. In this thesis, we, therefore, look into attacks that corrupt the measurements. The first part of the thesis is concerned with the feasibility of a worst-case sensor attack. The attacker's goal is to maximize its impact, while remaining undetected by an anomaly detector. The investigated worst-case attack strategy needs the exact controller state for its execution. Therefore, we start by looking into the feasibility of estimating the controller state by an attacker that has full model knowledge and access to all sensors. We show that an unstable controller prevents the attacker from estimating the controller state exactly and, therefore, makes the attack non-executable. Since unstable controllers come with their own issues, we propose a defense mechanism based on injecting uncertainty into the controller. Next, we examine the confidentiality of the anomaly detector. With access to the anomaly detector state, the attacker can design a more powerful attack. We show that, in the case of a detector with linear dynamics, the attacker is able to obtain an accurate estimate of the detector’s state. The second part of the thesis is concerned with the performance of anomaly detectors under the investigated attack in the first part. We use a previously proposed metric to compare the performance of a χ2, cumulative sum (CUSUM), and multivariate exponentially weighted moving average (MEWMA) detectors. This metric depends on the attack impact and average time between false alarms. For two different processes, we observe that the CUSUM and MEWMA detectors, which both have internal dynamics, can mitigate the attack impact more than the static χ2 detector. Since this metric depends on the attack impact, which is usually hard to determine, we then propose a new metric. The new metric depends on the number of sensors, and the size of an invariant set guaranteeing that the attack remains undetected. The new metric leads to similar results as the previously proposed metric, but is less dependent on the attack modeling. Finally, we formulate a Stackelberg game to tune the anomaly detector thresholds in a cost-optimal manner, where the cost depends on the number of false alarms and the impact an attack would cause. / Kritiska infrastrukturer, så som elnätet eller vattenförsörjningssystemet, är ryggraden i vårt moderna samhälle. Effektiviteten av kritiska infrastrukturerhar ökats genom integration med beräkningsenheter och kommunikationsnätverk, men detta har medfört att de också har blivit mer sårbara för cyberattacker. På grund av den underliggande fysikaliska processen kan dessa cyberattacker inte bara ha ekonomiska och ekologiska effekter, utan de kan också kosta människoliv. Flera rapporterade cyberattacker mot kritiska infrastrukturer visar att det är viktigt att skydda dem från dessa attacker. Kritiska infrastrukturer förlitar sig vanligtvis på noggranna sensormätningar för optimal prestanda. I denna avhandling undersöker vi därför attacker som korrumperar mätningar. Den första delen av avhandlingen handlar om genomförandet av en sensorattack i ett värstafallsscenario. Angriparens mål är att maximera verkan av attacken, medan den förblir oupptäckt av en feldetektor. Den undersökta värstafallstrategin behöver exakt information av regulatorns tillstånd för att kunna användas. Därför börjar vi med att titta på möjligheten att en angripare ska kunna uppskatta regulatorns tillstånd samtidigt som den känner till modellen och har tillgång till alla sensorer. Vi visar att en instabil regulator förhindrar angriparen från att exakt uppskatta regulatorns tillstånd och därmed förhindrar attacken. Eftersom instabila regulatorer introducerar andra problem, föreslår vi en försvarsmekanism baserad på injektion av osäkerhet i regulatorn. Därefter undersöker vi feldetektorns konfidentialitet. Med kännedom om feldetektorns tillstånd kan angriparen skapa en kraftfullare attack. Vi visar att angriparen kan få en noggrann uppskattning av detektorns tillstånd när detektorn har linjär dynamik. Den andra delen av avhandlingen behandlar feldetektorers prestanda medan de utsätts för de attacker som introducerades i första delen. Vi använder en tidigare föreslagen metrik för att jämföra prestandan av detektorer baserade på χ2-fördelningen, kumulativ summa (CUSUM), och multivariat exponentiellt viktat glidande medelvärde (MEWMA). Denna metrik beror på verkan av attacken och genomsnittlig tid mellan falska larm. Vi observerar att CUSUM- och MEWMA-detektorerna, där båda har intern dynamik, kan begränsa verkan av attacker bättre än vad den statiska χ2-detektorn kan för två olika processer. Eftersom denna metrik beror på attackens verkan, vilket vanligtvis är svårt att fastställa, föreslår vi en ny metrik. Den nya metriken beror på antalet sensorer och storleken på en invariant mängd som garanterar att attacken förblir oupptäckt. Den nya metriken leder till liknande resultat somden tidigare föreslagna metriken, men är mindre beroende av en modell av angriparen. Slutligen formulerar vi ett Stackelberg-spel för att ställa in trösklar för feldetektorn på ett kostnadsoptimalt sätt, där kostnaden beror på antalet falska larm och potentiell verkan av attacker.

Security and reliability are essential properties in Networked Control Systems (NCS), which are increasingly relevant in several important applications such as he process industry and electric power networks. The trend towards using non-proprietary and pervasive communication and information technology (IT) systems, such as the Internet and wireless communications, may result in NCS being vulnerable to cyber attacks. Traditional IT security does not consider the interdependencies between the physical components and the cyber realm of IT systems. Moreover, the control theoretic approach is not tailored to handle IT threats, focusing instead on nature-driven events. This thesis addresses the security and reliability of NCS, with a particular focus on power system control and supervision, contributing towards establishing a framework capable of analyzing and building NCS security. In our first contribution, the cyber security of the State Estimator (SE) in power networks is analyzed under malicious sensor data corruption attacks. The set of stealthy attacks bypassing current Bad Data Detector (BDD) schemes is characterized for the nonlinear least squares SE, assuming the attacker has accurate knowledge of a linearized model. This result is then extended to uncertain models using the geometric properties of the SE and BDD. Using the previous results, a security framework based on novel rational attack models is proposed, in which the minimum-effort attack policy is cast as a constrained optimization problem. The optimal attack cost is interpreted as a security metric, which can be used in the design of protective schemes to strengthen security. The features of the proposed framework are illustrated through simulation examples and experiments. As our second contribution, we analyze the behavior of the Optimal Power Flow (OPF) algorithmin the presence of stealthy sensor data corruption and the resulting consequences to the power network operation. In particular, we characterize the set of attacks that may lead the operator to apply the erroneous OPF recommendation and propose an analytical expression for the optimal solution of a simplified OPF problem with corrupted measurements. A novel impact-aware security metric is proposed based on these results, considering both the impact on the system and the attack cost. A small analytical example and numerical simulations are presented to illustrate and motivate our contributions. The third contribution considers the design of distributed schemes for fault detection and isolation in large-scale networks of second-order systems. The proposed approach is based on unknown input observers and exploits the networked structure of the system. Conditions are given on what local measurements should be available for the proposed scheme to be feasible. Infeasibility results with respect to available measurements and faults are also provided. In addition, methods to reduce the complexity of the proposed scheme are discussed, thus ensuring the scalability of the solution. Applications to power networks and robotic formations are presented through numerical examples. / QC 20111124